Fig. 2: The gameworld and the Conditioned Hallucinations (CH) task within it.

a The character’s avatar is located in the center of the screen in a which represents the gameworld. The yellow box, top left in a, represents a goal to be reached. The world is available for the player to explore and they may encounter computational tasks built into the environment as they seek to find paths towards the goal. Both task performance and exploration will be analyzed using computational models. As an example of this is the implementation of the visual version of the CH task: in the lower left of a, a dragon’s shadow is present. This is part of one implementation of the CH task (see b). Finally, in the red box on the lower right is an owl that may serve as a target for an ongoing attention task, should it be necessary to track player attention over time. b In this figure, we represent the traditional hierarchical gaussian filter (HGF) model as it pertains to the Conditioned Hallucinations (CH) game in the example gameworld. The version of the CH task demonstrated here is a visual conditioned hallucination as it is more intuitive to demonstrate in a static figure, but the auditory version of the CH task can easily be implemented as well. In the game, we would modify the traditional conditioned hallucinations task such that whenever a player hears a growl and sees a dragon shadow, they should dodge it. However, if they do not see the shadow, they should not jump. As has been done successfully in other iterations of the task, the participant learns to associate the shadow with the growl and reports seeing the shadow (by dodging) even when it is absent. In the HGF, there are three levels that form an agent’s perceptual model of the world in the game. Level 1 reflects the agent’s belief regarding the presence/absence of the dragon shadow on any given trial (trialwise P(V|A)). This is reflected in their decision to dodge or not. Level 2 reflects their belief that the dragon shadow is associated with the growl (overall P(V|A)). Finally, level 3 represents their belief in the volatility of the association between the growl and the shadow. Based on the participant’s decision to dodge or not, we can derive three important latent parameters that could act as behavior-based biomarkers of various psychiatric disorders: decision noise (β⁻¹), learning rate of the auditory-visual stimulus associations (ω), and weighting of prior expectations (ν).